Predicting adsorption ability of adsorbents at arbitrary sites for pollutants using deep transfer learning

Xirong Lin,Haikuo Zhang,Tianli Han,Jinyun Liu,Jiahao Ren,Jinjin Li,Zhilong Wang

doi:10.1038/s41524-021-00494-9

Xirong Lin, Haikuo Zhang + Show 5 more

Open Access

https://doi.org/10.1038/s41524-021-00494-9

Copy DOI

Abstract

Accurately evaluating the adsorption ability of adsorbents for heavy metal ions (HMIs) and organic pollutants in water is critical for the design and preparation of emerging highly efficient adsorbents. However, predicting adsorption capabilities of adsorbents at arbitrary sites is challenging, with currently unavailable measuring technology for active sites and the corresponding activities. Here, we present an efficient artificial intelligence (AI) approach to predict the adsorption ability of adsorbents at arbitrary sites, as a case study of three HMIs (Pb(II), Hg(II), and Cd(II)) adsorbed on the surface of a representative two-dimensional graphitic-C3N4. We apply the deep neural network and transfer learning to predict the adsorption capabilities of three HMIs at arbitrary sites, with the predicted results of Cd(II) > Hg(II) > Pb(II) and the root-mean-squared errors less than 0.1 eV. The proposed AI method has the same prediction accuracy as the ab initio DFT calculation, but is millions of times faster than the DFT to predict adsorption abilities at arbitrary sites and only requires one-tenth of datasets compared to training from scratch. We further verify the adsorption capacity of g-C3N4 towards HMIs experimentally and obtain results consistent with the AI prediction. It indicates that the presented approach is capable of evaluating the adsorption ability of adsorbents efficiently, and can be further extended to other interdisciplines and industries for the adsorption of harmful elements in aqueous solution.

Highlights

Recent studies have shown that when artificial intelligence (AI) meets material design and discovery, it means reducing the time and cost going from lab to practical applications by greatly improving the research efficiency[1,2,3]
The root-mean-squared errors (RMSEs) of 0.1 eV obtained from the prediction model by only a few hundred density functional theory (DFT) calculations is a remarkable achievement, which provides a powerful guarantee for the statistical prediction of adsorption capacity of materials to Heavy metal ions (HMIs)
We proposed an AI approach to evaluate the adsorption ability of adsorbent toward HMIs at arbitrary sites accurately describe the interaction between HMIs and g-C3N4 substrate, the DFT-D3 method[47] was employed, which considers the van der Waals interaction

Summary

INTRODUCTION

Recent studies have shown that when artificial intelligence (AI) meets material design and discovery, it means reducing the time and cost going from lab to practical applications by greatly improving the research efficiency[1,2,3]. The RMSE of 0.1 eV obtained from the prediction model by only a few hundred DFT calculations is a remarkable achievement, which provides a powerful guarantee for the statistical prediction of adsorption capacity of materials to HMIs. As expected, based on the structural descriptor of LEM, we can fleetly predict the adsorption energy of Pb(II) at the arbitrary site with an accuracy of 0.051 eV for the 1000. To clarify the rationality and accuracy of the TL method in processing small datasets, Fig. 4 shows the performance comparison of trained models from scratch (FS) and transfer learning (TL) in each iteration for Hg(II)/g-C3N4 and Cd(II)/g-C3N4, based on 700 single-point adsorption energies by DFT calculation. We expect that it will be significant to further study the practical application of the presented ML method, which would be an important research direction

DISCUSSION

METHODS

Findings

CODE AVAILABILITY